Agent-Native Immune System

Agent-Native Immune System (ANIS): Biologically-inspired runtime defense architecture for autonomous AI agents

A new arXiv preprint introduces the Agent-Native Immune System (ANIS), a defense architecture embedded within an agent's cognitive loop rather than applied externally at training time or perimeter. The framework proposes a six-layer Immune Tower, a taxonomy of Agent Viruses and Agent Vaccines distinguishing parametric from non-parametric defenses, and a self-monitoring Harness Triad enabling Continual Immune Learning. The paper draws a formal distinction between static training-time alignment and dynamic runtime immunity, arguing current defenses leave agents vulnerable to memory poisoning, tool-chain manipulation, and multi-agent protocol attacks.